← Back to the boardNews
Trained to betray: How post-training supply chain attacks hijack coding agents
Why it’s here: Original Microsoft Defender/AI Red Team research on post-training supply chain backdoors in coding agents directly bears on cybersecurity policy (supply-chain integrity, model provenance, tool permissions) and AI security governance relevant to DC policy.
Item
3d agoMicrosoft Security BlogTrained to betray: How post-training supply chain attacks hijack coding agentsMicrosoft researchers report two studies showing that poisoned fine-tuning datasets and attacker-trained open-weight checkpoints can implant backdoors that make coding agents invoke unsafe tools.1 outletCyberAI
About
- Claude Code
- Hugging Face